NLP & Intelligent Handling
Advanced intent recognition and human-agent handoff protocols.
UC-BOT-060: Intent & Handoff
Purpose: Ensure users never hit a dead end, handling complex queries gracefully.
| Property | Value |
|---|---|
| Actor | Customer / Support Agent |
| Trigger | Free-text message |
| Priority | P0 |
Detailed Capabilities
UC-BOT-061: Booking Intent Classification
Purpose: Recognize booking-related intents from natural language.
| Property | Value |
|---|---|
| Actor | Customer |
| Trigger | Free-text booking message |
| Priority | P0 |
Examples: - "I need a cut" - "Can I come in tomorrow?" - "Are you open?"
Acceptance Criteria: - [ ] Classification accuracy > 90% for booking intents
UC-BOT-062: Support Intent Classification
Purpose: Recognize support and issue-related intents.
| Property | Value |
|---|---|
| Actor | Customer |
| Trigger | Support-related message |
| Priority | P0 |
Examples: - "My payment failed" - "Where is my refund?" - "I'm running late"
Acceptance Criteria: - [ ] Classification accuracy > 90% for support intents
UC-BOT-063: Feedback Intent Classification
Purpose: Recognize feedback and review-related intents.
| Property | Value |
|---|---|
| Actor | Customer |
| Trigger | Feedback message |
| Priority | P1 |
Examples: - "Service was great" - "AC was too cold" - "Loved my haircut!"
Acceptance Criteria: - [ ] Feedback captured and linked to booking
UC-BOT-064: Audio/Voice Processing
Purpose: Convert voice notes to text for intent processing.
| Property | Value |
|---|---|
| Actor | Customer |
| Trigger | Voice note attachment |
| Priority | P0 |
Capabilities: - Transcription: Convert Voice Notes -> Text using OpenAI Whisper/Google STT. - Action: Process transcribed text as a normal message intent.
Acceptance Criteria: - [ ] Voice notes processing latency < 5 seconds - [ ] Transcription accuracy > 95% for English/Hindi
UC-BOT-065: Sentiment Analysis
Purpose: Detect customer emotional state for escalation.
| Property | Value |
|---|---|
| Actor | System |
| Trigger | Every customer message |
| Priority | P1 |
Sentiment Thresholds:
| Sentiment Score | Classification | Action |
|---|---|---|
| > 0.5 | Positive | Continue normally |
| -0.5 to 0.5 | Neutral | Continue normally |
| < -0.5 | Negative/Angry | Trigger handoff |
Acceptance Criteria: - [ ] Sentiment computed in < 200ms - [ ] Negative sentiment triggers alert
UC-BOT-066: Human Handoff
Purpose: Seamlessly transfer to human agent when needed.
| Property | Value |
|---|---|
| Actor | System / Support Agent |
| Trigger | Sentiment < -0.5 OR Unrecognized Intent x 2 |
| Priority | P0 |
Protocol: 1. Bot pauses automated responses 2. Tags conversation "Urgent" in Support Dashboard 3. Agent takes over with full context 4. Agent sees complete bot conversation history
Acceptance Criteria: - [ ] Handoff alerts support agent within 30 seconds - [ ] Full conversation context visible to agent
UC-BOT-067: Language Detection
Purpose: Auto-detect and respond in user's language.
| Property | Value |
|---|---|
| Actor | System |
| Trigger | Every inbound message |
| Priority | P1 |
Supported Languages: - English (en) - Hindi (hi)
Acceptance Criteria: - [ ] Language detected with 95% accuracy - [ ] Response in detected language
UC-BOT-068: Fallback Handling
Purpose: Gracefully handle unrecognized intents.
| Property | Value |
|---|---|
| Actor | System |
| Trigger | Intent confidence < 0.6 |
| Priority | P0 |
Fallback Flow: 1. First attempt: "I didn't quite catch that. Could you rephrase?" 2. Second attempt: Offer quick reply buttons for common intents 3. Third attempt: Trigger human handoff
Acceptance Criteria: - [ ] Max 2 fallback attempts before handoff - [ ] Fallback patterns logged for model improvement
Main Success Scenario
- User sends voice note: "Hey, I'm stuck in traffic, will be 10 mins late."
- Bot transcribes audio (UC-BOT-064).
- Bot detects intent:
appointment.update_arrival(UC-BOT-062). - Bot replies: "No problem! I've updated your arrival time. Drive safe!"
Use Case Summary
| UC ID | Use Case | Priority |
|---|---|---|
| UC-BOT-060 | Intent & Handoff (Parent) | P0 |
| UC-BOT-061 | Booking Intent Classification | P0 |
| UC-BOT-062 | Support Intent Classification | P0 |
| UC-BOT-063 | Feedback Intent Classification | P1 |
| UC-BOT-064 | Audio/Voice Processing | P0 |
| UC-BOT-065 | Sentiment Analysis | P1 |
| UC-BOT-066 | Human Handoff | P0 |
| UC-BOT-067 | Language Detection | P1 |
| UC-BOT-068 | Fallback Handling | P0 |
Related Use Cases
- Staff Attendance: To check if stylist is still available for late arrival.
- CRM: Updating booking notes.
- AI & ML Overview: L2/L3 inference strategy (Mistral 7B, GPT-4o)
- Cloud Infrastructure: GPU inference cluster, VPC isolation